I’ve noticed this water narrative coming up in many other unrelated stories. When the protests against the Keystone XL pipeline were at their height, it was actually water that was the focal issue rather than oil itself. (The slogan of the Standing Rock protests was “water is life”.)
At around that time, I had noticed that Nestle was becoming one of the most hated companies on Reddit, because of various facilities to bottle water. People got extremely upset at one facility in Michigan, where there is nothing like a water shortage, because it somehow seemed insensitive to bottle water in Michigan at a time when the Flint water infrastructure was leaking lead. People didn’t quite understand that the issue was lead - they somehow thought Flint had a water shortage, and assumed Nestle was somehow making it worse.
As my old grad school mentor used to say to me: "You cannot reason your way out of something you were not reasoned *into*"
But I wonder, do you think critics "already feel" negatively towards AI and just cannot accurately pinpoint where the feeling is coming from and therefore latch on to the easy-to-adopt water argument? Meaning, do you think the water issue is making some people anti-AI or is it that anti-AI people are using water as the issue that everyone can agree on as a case against it?
Good question. I think it's mostly the latter: water is a safe bet for anti-AI people! But there's a bit of the former (people who don't know much about AI will become wary of it when they hear about the water thing).
Thanks for this. 2026 promises to be another year of great posts. Your psychological autopsy is so wide-ranging that I’m inclined to sketch a massive mind map just to keep thinking along these many lines of inquiry . . .
I guess the correct (pragmatic) answer is to join the communities where this already happens (e.g. rationalists), but I don't think it's possible to impose that change into the broader population
This issue has driven me nuts since it first went through. The bigger mistake is that they mistook 'flow' for 'consumption.' Most water flows through a cooling system at a rate, say 100 gallons a minute. Therefore, if you asked ChatGPT 1,000 question a minute they said each question 'consumed' 1/10th of a gallon. But it's not like drinking water and pissing it out, it flows back to the holding system, cools, and cycles back. The evaporation is something they want to avoid because water isn't cheap. The biggest mistake they made was equating flow with consumption.
I get the concern's over resource use, but to me that concern is better directed at turning the world into a server centre, or worse, going into space to pollute that too.
I even find a better criticism that a lot of investment that is directed to data centers could be directed to other things than water. Water is the easiest to make and possibly the worst/least correct.
Thank you for this. As is often the case, discovering new ways of thinking about things/seeing the problem, is more enlightening that discovering new facts
Concrete ask of those who want to advance arguments against AI based on resource consumption: drop water, map your concerns to energy.
I get that it isn't as pure a symbol, but the contention and possible carbon impacts of AI energy use aren't so dismissible that those who want to ally with you in pushing back against the AI race feel obliged to disclaim and distance themselves from you.
Great post. Intuitively, the ‘AI and data centres use too much water and electricity’ concern is unfounded because presumably it is economically feasible to buy the water and electricity to run the data centres, and if it wasn’t then the price of using AI would be prohibitively expensive and therefore the technology would not be commercially available
Yes, but with "too much" they don't mean literally "possible" but "desirable". It turns out that water usage is not that high as to be undesirable in most cases (there might be exceptions where careful handling of supply is critical), except for those people who deem AI inherently undesirable.
Yep. Water and electricity infrastructure responds to supply and demand. If there are customers willing to pay, the utility providers will build more capacity. In Australia water is very precious, but that is mostly due to the way the state governments along the Murray Darling basin agree to distribute the available water as it moves downstream, making it very scarce for settlements at the mouth of the river. The amount of water available for agriculture is obscene for the tiny price many of the farmers pay. It would dwarf the amount of water used by data centres. Urban utility customers pay exorbitant prices per ML, while farmers pay practically nothing, due to historical arrangements and political lobbying
It is far from my biggest issue re racing to superintelligence (that would be the "kills everyone" concern) but especially in the US the energy resource concern has some merit and nuance, while the water resource concern is embarrassing.
(The industry is structured such that existing consumers do share the cost of investing in enormous new supply, and it does increase demand for carbon based energy sources.)
Thought-provoking take! This explains *why* the water narrative sticks far better than another data-center chart ever could. Nuance vs symbols is the real battle here
Bubble, perhaps, but I've seen dozens of debunkings of the LLM water use issue months before I've seen anyone actually use it as a straight criticism. And in the rare cases I see it used straight, it's never more than a brief throwaway remark.
I think the explanation may be much simpler and people are essentially trolling. As in, they see how the issue riles the pro-AI people up and reasonably conclude it's a good way to piss those people off.
The comparison with the impact of food is the apex, in my humble opinion. Most of the water impact a person has during life comes from food consumption. Even so, that information is buried under a lot of advertising and misinformation.
If people really cared about water, changing the diet would be the first step.
My only nitpicks would be to consider potable water usage once we have built out the un-numbered data centers in various phases of buildout, while also considering that in the US at least, with its outdated grid, these things will be powered by fracking and coal. So harm to the water tables is firmly a proven outcome
Not saying youre incorrect but I will always stand with the locals who say data centers are deleterious to their communities
Yes, I think it's not wise to dismiss the criticisms either away, rather should always be investigated. But once that happens, let's not ignore the data if it turns out the water issue is not that important (it could still be in specific locations, though!)
Ok, we can have surveys and we can listen to the people who are already in places where a data center was forced on them
But I have to ask, what are we solving for? All the researchers I know who use Ai agents run them on device, it’s a matter of IP and security. So what do you think we are solving for with un-numbered data centers?
I think there's a race to build datacenters because the AI industry thinks there's demand for it. I think generative AI has not yet proved worthy of such investment but that's my opinion. Some people run LLMs on device but not most people. We use cloud services. I think it's a mistake and the industry is over-building, but that's another topic (I've written about it as well!)
I’ve noticed this water narrative coming up in many other unrelated stories. When the protests against the Keystone XL pipeline were at their height, it was actually water that was the focal issue rather than oil itself. (The slogan of the Standing Rock protests was “water is life”.)
At around that time, I had noticed that Nestle was becoming one of the most hated companies on Reddit, because of various facilities to bottle water. People got extremely upset at one facility in Michigan, where there is nothing like a water shortage, because it somehow seemed insensitive to bottle water in Michigan at a time when the Flint water infrastructure was leaking lead. People didn’t quite understand that the issue was lead - they somehow thought Flint had a water shortage, and assumed Nestle was somehow making it worse.
Very lenghty but an excellent read.
🙏🙏🙏
As my old grad school mentor used to say to me: "You cannot reason your way out of something you were not reasoned *into*"
But I wonder, do you think critics "already feel" negatively towards AI and just cannot accurately pinpoint where the feeling is coming from and therefore latch on to the easy-to-adopt water argument? Meaning, do you think the water issue is making some people anti-AI or is it that anti-AI people are using water as the issue that everyone can agree on as a case against it?
Good question. I think it's mostly the latter: water is a safe bet for anti-AI people! But there's a bit of the former (people who don't know much about AI will become wary of it when they hear about the water thing).
Thanks for this. 2026 promises to be another year of great posts. Your psychological autopsy is so wide-ranging that I’m inclined to sketch a massive mind map just to keep thinking along these many lines of inquiry . . .
Thank you Mary, and happy new year (much more to come!)
If the problem is ultimately people seeking in-group status rather than the truth, then we need to make truth-seeking high-status.
Any ideas on how to do this?
I guess the correct (pragmatic) answer is to join the communities where this already happens (e.g. rationalists), but I don't think it's possible to impose that change into the broader population
This issue has driven me nuts since it first went through. The bigger mistake is that they mistook 'flow' for 'consumption.' Most water flows through a cooling system at a rate, say 100 gallons a minute. Therefore, if you asked ChatGPT 1,000 question a minute they said each question 'consumed' 1/10th of a gallon. But it's not like drinking water and pissing it out, it flows back to the holding system, cools, and cycles back. The evaporation is something they want to avoid because water isn't cheap. The biggest mistake they made was equating flow with consumption.
Yeah, I think this is a big part of it. Would this conceptual shift work? I'm not sure, but it seems like it's simple enough that it could work
I get the concern's over resource use, but to me that concern is better directed at turning the world into a server centre, or worse, going into space to pollute that too.
I even find a better criticism that a lot of investment that is directed to data centers could be directed to other things than water. Water is the easiest to make and possibly the worst/least correct.
Thank you for this. As is often the case, discovering new ways of thinking about things/seeing the problem, is more enlightening that discovering new facts
Thanks for reading Nick
Concrete ask of those who want to advance arguments against AI based on resource consumption: drop water, map your concerns to energy.
I get that it isn't as pure a symbol, but the contention and possible carbon impacts of AI energy use aren't so dismissible that those who want to ally with you in pushing back against the AI race feel obliged to disclaim and distance themselves from you.
Excellent post as always...
Thank you Kenneth!
We don’t always agree and that is ok i still love your work and your passion. You have always done good work but i will keep poking you lol.
It's good to be poked, keeps me sharp!
Great post. Intuitively, the ‘AI and data centres use too much water and electricity’ concern is unfounded because presumably it is economically feasible to buy the water and electricity to run the data centres, and if it wasn’t then the price of using AI would be prohibitively expensive and therefore the technology would not be commercially available
Yes, but with "too much" they don't mean literally "possible" but "desirable". It turns out that water usage is not that high as to be undesirable in most cases (there might be exceptions where careful handling of supply is critical), except for those people who deem AI inherently undesirable.
Yep. Water and electricity infrastructure responds to supply and demand. If there are customers willing to pay, the utility providers will build more capacity. In Australia water is very precious, but that is mostly due to the way the state governments along the Murray Darling basin agree to distribute the available water as it moves downstream, making it very scarce for settlements at the mouth of the river. The amount of water available for agriculture is obscene for the tiny price many of the farmers pay. It would dwarf the amount of water used by data centres. Urban utility customers pay exorbitant prices per ML, while farmers pay practically nothing, due to historical arrangements and political lobbying
It is far from my biggest issue re racing to superintelligence (that would be the "kills everyone" concern) but especially in the US the energy resource concern has some merit and nuance, while the water resource concern is embarrassing.
(The industry is structured such that existing consumers do share the cost of investing in enormous new supply, and it does increase demand for carbon based energy sources.)
Thought-provoking take! This explains *why* the water narrative sticks far better than another data-center chart ever could. Nuance vs symbols is the real battle here
Right, all that data and the story remains the same (although I think it does fundamental work still; without the data I could not write this!)
Bubble, perhaps, but I've seen dozens of debunkings of the LLM water use issue months before I've seen anyone actually use it as a straight criticism. And in the rare cases I see it used straight, it's never more than a brief throwaway remark.
I think the explanation may be much simpler and people are essentially trolling. As in, they see how the issue riles the pro-AI people up and reasonably conclude it's a good way to piss those people off.
If bubble, then it's a good bubble to be in!
In our age the truth doesn't matter as much as narrative does.
Really, really good content.
The comparison with the impact of food is the apex, in my humble opinion. Most of the water impact a person has during life comes from food consumption. Even so, that information is buried under a lot of advertising and misinformation.
If people really cared about water, changing the diet would be the first step.
I find it funny how frequently the death or torture of puppies is used to suggest ‘this thing is objectively morally bad’ in political satire
You’ve laid out a great argument
My only nitpicks would be to consider potable water usage once we have built out the un-numbered data centers in various phases of buildout, while also considering that in the US at least, with its outdated grid, these things will be powered by fracking and coal. So harm to the water tables is firmly a proven outcome
Not saying youre incorrect but I will always stand with the locals who say data centers are deleterious to their communities
Yes, I think it's not wise to dismiss the criticisms either away, rather should always be investigated. But once that happens, let's not ignore the data if it turns out the water issue is not that important (it could still be in specific locations, though!)
Ok, we can have surveys and we can listen to the people who are already in places where a data center was forced on them
But I have to ask, what are we solving for? All the researchers I know who use Ai agents run them on device, it’s a matter of IP and security. So what do you think we are solving for with un-numbered data centers?
I think there's a race to build datacenters because the AI industry thinks there's demand for it. I think generative AI has not yet proved worthy of such investment but that's my opinion. Some people run LLMs on device but not most people. We use cloud services. I think it's a mistake and the industry is over-building, but that's another topic (I've written about it as well!)
Caiste en mi trampa!
Let’s not argue for a thing that doesnt solve the resource cost, just because the resource cost is not articulated properly